In private equity, we evaluate every deal the same way. What’s the risk-adjusted return? What’s the entry multiple? What’s the path to exit? What could kill this investment?
After spending over a decade evaluating hundreds of deals at a PE firm, I started noticing something: the same framework applies almost perfectly to enterprise AI decisions. The “build vs buy” question isn’t a technology question. It’s an investment question. And most organizations are making it with the wrong toolkit.
The Framework: Five Dimensions
1. Total Cost of Ownership (TCO)
Buy: Monthly SaaS subscription × number of users × years. Looks cheap in Year 1. But enterprise AI pricing is evolving fast — per-seat, per-query, per-token, per-API-call. I’ve seen projections where a cloud AI tool costs more than the salary of the team it was supposed to replace, once you hit real usage volumes.
Build: Hardware (servers, GPUs if needed), talent (or time, if you build in-house), maintenance, and power. Higher upfront cost, but the marginal cost of an additional query is essentially zero.
The PE lens: This is a classic operating leverage question. Build has high fixed costs and low variable costs. Buy has low fixed costs and high variable costs. The crossover point depends on usage volume. If you’re processing 50 documents a month, buy. If you’re processing 5,000, run the numbers carefully.
2. Time to Value
Buy: Weeks to deploy. Vendor handles the infrastructure. You configure, integrate, train your team.
Build: Months to a working prototype. More months to production-grade reliability.
But here’s the catch: Buy comes with customization debt. The vendor’s system works for the general case. Your use case is never the general case. The time you save in deployment, you spend in workarounds, data transformation, and “that feature isn’t available yet” conversations with support.
The PE lens: In fund investing, we distinguish between time-to-first-close and time-to-full-deployment. A quick first close means nothing if you can’t deploy the capital effectively. Same principle — quick deployment means nothing if adoption stalls because the tool doesn’t fit your workflows.
3. Data Sovereignty
Buy: Your data goes to the vendor’s servers. They promise encryption, access controls, SOC 2 compliance. But at the end of the day, your proprietary information is sitting on someone else’s infrastructure, governed by their security practices and their jurisdiction’s laws.
Build: Your data never leaves your building. Full stop. You control access, logging, and retention. Your compliance team sleeps at night.
The PE lens: This is a risk-adjusted return question. The expected return of cloud AI might be higher (faster deployment, better models). But the tail risk — a data breach involving confidential deal documents, investor information, or legal agreements — is catastrophic. In PE, we call this “picking up pennies in front of a steamroller.” For regulated industries, on-premise is the risk-adjusted winner.
4. Moat and Differentiation
Buy: You’re using the same tool as your competitors. Your advantage is execution and adoption speed, not the technology itself. When everyone has the same AI assistant, the playing field is level.
Build: You own the system. You can train it on your institutional knowledge — years of deal memos, diligence reports, negotiation outcomes. Over time, the system gets better because it learns from your data. That’s a compounding advantage no competitor can replicate by swiping a credit card.
The PE lens: In fund investing, we look for durable competitive advantages — what Warren Buffett calls “moats.” A SaaS subscription is not a moat. A proprietary AI system trained on a decade of your institutional knowledge is.
5. Exit Optionality
Buy: Switching costs compound over time. Your workflows integrate around the vendor’s API. Your team trains on the vendor’s interface. Your data accumulates in the vendor’s format. Two years in, switching isn’t a technology decision — it’s an organizational disruption.
Build: You own the IP. You can upgrade components independently (swap the language model, change the embedding approach, restructure the retrieval pipeline). No vendor lock-in. No pricing renegotiations where the leverage has shifted.
The PE lens: We always evaluate an investment with the exit in mind. What does the exit look like in 3-5 years? With buy, you’re locked in and your bargaining power decreases each year. With build, your options expand as the technology improves and open-source alternatives mature.
The Verdict: It Depends on Your Thesis
Just like in fund investing, there’s no universal answer. The right decision depends on your organizational context:
Buy when:
- Your AI use case is generic (email summarization, meeting notes, general Q&A)
- Data sensitivity is low
- You need to move fast and the tool fits 80%+ of your needs
- You do not need to own or deeply customize the underlying system
Build when:
- Your use case involves proprietary or sensitive data
- You’re in a regulated industry (finance, legal, healthcare, real estate)
- Your competitive advantage depends on institutional knowledge
- You plan to scale AI across multiple use cases over time
- You want to own and control how the system evolves
The hybrid approach (often the smartest): Build your core — the system that touches your sensitive data and creates competitive advantage. Buy the periphery — generic productivity tools, email integration, scheduling.
The Real Insight
Most organizations approach the build-vs-buy question as a technology decision and delegate it to their IT department. That’s like asking your construction contractor to decide whether to build or buy a property. They’ll give you a construction answer.
This is fundamentally a capital allocation decision. It requires understanding total cost of ownership, risk-adjusted returns, competitive moats, and exit optionality. It requires someone who can talk to the board in the language of investment returns and talk to the engineering team in the language of system architecture.
The best AI strategy doesn’t come from your smartest engineer or your most experienced vendor consultant. It comes from the person who can evaluate it like an investment — because that’s exactly what it is.